Tech Bro’s AI-Generated FPS Is a Complete Disaster
▼ Summary
– A viral video shows a bizarre AI-generated game that appears to be a glitchy, on-rails shooter with reality-warping visuals.
– The clip was shared by HyperWrite CEO Matt Shuman as a teaser for an upcoming open-source, community-driven project.
– The game features a choose-your-own-adventure style where the AI generates scenes based on periodic player prompt selections.
– Visuals include melting environments, illogical object placements, and nonsensical elements like multiple “Uptoon” transportation signs.
– Shuman previously faced fraud accusations for promoting bogus benchmarks related to his AI company’s language model.
The recent unveiling of an AI-generated first-person shooter has ignited a firestorm of reactions online, with many viewers describing the experience as a surreal and unsettling descent into digital chaos. Shared by HyperWrite CEO Matt Shuman, the brief video clip was presented with the optimistic caption, “AI games are going to be amazing (sound on).” The actual footage, however, paints a starkly different picture, resembling a glitch-ridden, on-rails shooter that is both disturbing to watch and strangely compelling.
Shuman promoted the clip on the social platform X, framing it as a preview for an upcoming, completely open-source project intended for community development. At first glance, the game appears to be a third-person shooter, but a row of periodic prompt choices at the screen’s bottom reveals its true nature as a choose-your-own-adventure experience. In this format, the AI dynamically constructs each subsequent scene based on the player’s selections.
What unfolds is a visual disaster. The game’s reality seems to crumble and fold in on itself, creating a series of jarring, Matrix-like transitions. A bus platform inexplicably melts into a subway station, cars appear underground, and HVAC systems are stacked unnaturally on top of one another. The environment looks like a violent collage of mismatched film reels, with transportation signs consistently displaying the nonsensical word “Uptoon.”
The character’s actions are equally broken. Any attempt by the SWAT-style avatar to perform basic functions, such as firing a weapon, throwing a grenade, or descending a ladder, results in the kind of grotesque glitch typically seen in failed AAA game launches. A particularly absurd moment occurs near the clip’s end when the character is instructed to find cover from an enemy helicopter, despite already being safely inside a sewer beneath the street.
The video concludes by cutting to a random pile of leftover PVC pipe cuttings, an ending that feels entirely disconnected from the preceding action. The overall presentation bears less resemblance to a professional game reveal and more to the spammy, brain-melting mobile game ads that flood social media feeds. It’s a spectacle that holds your attention out of morbid curiosity, yet offers no coherent narrative or meaning behind its algorithmically generated parade of unrelated images.
This is not the first time Shuman has been at the center of controversy. A year ago, tech publication VentureBeat published a glowing profile announcing his company’s new large language model, Reflection 70B. The article praised the model, which is based on Meta’s open-source Llama 3.1-70B Instruct, for leveraging a novel error self-correction technique and demonstrating superior performance on third-party benchmarks. However, just days after that announcement, Shuman faced accusations of fraud, with critics alleging that the impressive benchmarks he was promoting were entirely bogus.
(Source: Kotaku)